Reasoning on Gene Ontology Networks Predicts Novel Protein Annotations

Careful examination of the existing evidence underpins formulation of novel hypotheses and experimental design. However, each year the state of the ground truth becomes more difficult to approximate due to the rapid proliferation of primary literature and associated databases. With over a million new articles added to MEDLINE annually, mastering even the subfield literature has become difficult. To address this problem, algorithmic approaches have been developed to automatically collect, integrate, and even reason on data. For instance, in 2014 the KnIT method condensed the entirety of the human kinome literature into a single graph, and then reasoned on it to predict novel kinases that target a critical cancer suppressor protein (1). While these methods are powerful, they rely on primary literature as their main source of data, and therefore ignore structured data contained in the multitude of domain databases.